Dominik Moritz

Fred Hohman, Mary Beth Kery, Donghao Ren et al. · apple-ml, cmu

On-device machine learning (ML) promises to improve the privacy, responsiveness, and proliferation of new, intelligent user experiences by moving ML computation onto everyday personal devices. However, today's large ML models must be drastically compressed to run efficiently on-device, a hurtle that requires deep, yet currently niche expertise. To engage the broader human-centered ML community in on-device ML experiences, we present the results from an interview study with 30 experts at Apple that specialize in producing efficient models. We compile tacit knowledge that experts have developed through practical experience with model compression across different hardware platforms. Our findings offer pragmatic considerations missing from prior work, covering the design process, trade-offs, and technical strategies that go into creating efficient models. Finally, we distill design recommendations for tooling to help ease the difficulty of this work and bring on-device ML into to more widespread practice.

Samantha Robertson, Zijie J. Wang, Dominik Moritz et al. · apple-ml, cmu

Machine learning (ML) models can fail in unexpected ways in the real world, but not all model failures are equal. With finite time and resources, ML practitioners are forced to prioritize their model debugging and improvement efforts. Through interviews with 13 ML practitioners at Apple, we found that practitioners construct small targeted test sets to estimate an error's nature, scope, and impact on users. We built on this insight in a case study with machine translation models, and developed Angler, an interactive visual analytics tool to help practitioners prioritize model improvements. In a user study with 7 machine translation experts, we used Angler to understand prioritization practices when the input space is infinite, and obtaining reliable signals of model quality is expensive. Our study revealed that participants could form more interesting and user-focused hypotheses for prioritization by analyzing quantitative summary statistics and qualitatively assessing data by reading sentences.

Xinyi Zheng, Ryan A. Rossi, Nesreen Ahmed et al. · apple-ml, cmu

The rapid development of network science and technologies depends on shareable datasets. Currently, there is no standard practice for reporting and sharing network datasets. Some network dataset providers only share links, while others provide some contexts or basic statistics. As a result, critical information may be unintentionally dropped, and network dataset consumers may misunderstand or overlook critical aspects. Inappropriately using a network dataset can lead to severe consequences (e.g., discrimination) especially when machine learning models on networks are deployed in high-stake domains. Challenges arise as networks are often used across different domains (e.g., network science, physics, etc) and have complex structures. To facilitate the communication between network dataset providers and consumers, we propose network report. A network report is a structured description that summarizes and contextualizes a network dataset. Network report extends the idea of dataset reports (e.g., Datasheets for Datasets) from prior work with network-specific descriptions of the non-i.i.d. nature, demographic information, network characteristics, etc. We hope network reports encourage transparency and accountability in network research and development across different fields.

Angie Boggust, Venkatesh Sivaraman, Yannick Assogba et al. · apple-ml, cmu

To deploy machine learning models on-device, practitioners use compression algorithms to shrink and speed up models while maintaining their high-quality output. A critical aspect of compression in practice is model comparison, including tracking many compression experiments, identifying subtle changes in model behavior, and negotiating complex accuracy-efficiency trade-offs. However, existing compression tools poorly support comparison, leading to tedious and, sometimes, incomplete analyses spread across disjoint tools. To support real-world comparative workflows, we develop an interactive visual system called Compress and Compare. Within a single interface, Compress and Compare surfaces promising compression strategies by visualizing provenance relationships between compressed models and reveals compression-induced behavior changes by comparing models' predictions, weights, and activations. We demonstrate how Compress and Compare supports common compression analysis tasks through two case studies, debugging failed compression on generative language models and identifying compression artifacts in image classification models. We further evaluate Compress and Compare in a user study with eight compression experts, illustrating its potential to provide structure to compression workflows, help practitioners build intuition about compression, and encourage thorough analysis of compression's effect on model behavior. Through these evaluations, we identify compression-specific challenges that future visual analytics tools should consider and Compress and Compare visualizations that may generalize to broader model comparison tasks.

Aspen Hopkins, Fred Hohman, Luca Zappella et al. · apple-ml, cmu

Lack of diversity in data collection has caused significant failures in machine learning (ML) applications. While ML developers perform post-collection interventions, these are time intensive and rarely comprehensive. Thus, new methods to track & manage data collection, iteration, and model training are necessary for evaluating whether datasets reflect real world variability. We present designing data, an iterative approach to data collection connecting HCI concepts with ML techniques. Our process includes (1) Pre-Collection Planning, to reflexively prompt and document expected data distributions; (2) Collection Monitoring, to systematically encourage sampling diversity; and (3) Data Familiarity, to identify samples that are unfamiliar to a model using density estimation. We apply designing data to a data collection and modeling task. We find models trained on ''designed'' datasets generalize better across intersectional groups than those trained on similarly sized but less targeted datasets, and that data familiarity is effective for debugging datasets.

Michelle S. Lam, Fred Hohman, Dominik Moritz et al. · apple-ml, cmu

AI policy sets boundaries on acceptable behavior for AI models, but this is challenging in the context of large language models (LLMs): how do you ensure coverage over a vast behavior space? We introduce policy maps, an approach to AI policy design inspired by the practice of physical mapmaking. Instead of aiming for full coverage, policy maps aid effective navigation through intentional design choices about which aspects to capture and which to abstract away. With Policy Projector, an interactive tool for designing LLM policy maps, an AI practitioner can survey the landscape of model input-output pairs, define custom regions (e.g., "violence"), and navigate these regions with if-then policy rules that can act on LLM outputs (e.g., if output contains "violence" and "graphic details," then rewrite without "graphic details"). Policy Projector supports interactive policy authoring using LLM classification and steering and a map visualization reflecting the AI practitioner's work. In an evaluation with 12 AI safety experts, our system helps policy designers craft policies around problematic model behaviors such as incorrect gender assumptions and handling of immediate physical safety threats.

Racquel Fygenson, Kazi Jawad, Isabel Li et al. · apple-ml, cmu

Reconstruction of 3D scenes from 2D images is a technical challenge that impacts domains from Earth and planetary sciences and space exploration to augmented and virtual reality. Typically, reconstruction algorithms first identify common features across images and then minimize reconstruction errors after estimating the shape of the terrain. This bundle adjustment (BA) step optimizes around a single, simplifying scalar value that obfuscates many possible causes of reconstruction errors (e.g., initial estimate of the position and orientation of the camera, lighting conditions, ease of feature detection in the terrain). Reconstruction errors can lead to inaccurate scientific inferences or endanger a spacecraft exploring a remote environment. To address this challenge, we present VECTOR, a visual analysis tool that improves error inspection for stereo reconstruction BA. VECTOR provides analysts with previously unavailable visibility into feature locations, camera pose, and computed 3D points. VECTOR was developed in partnership with the Perseverance Mars Rover and Ingenuity Mars Helicopter terrain reconstruction team at the NASA Jet Propulsion Laboratory. We report on how this tool was used to debug and improve terrain reconstruction for the Mars 2020 mission.

Tom Gunter, Zirui Wang, Chong Wang et al.

We present foundation language models developed to power Apple Intelligence features, including a ~3 billion parameter model designed to run efficiently on devices and a large server-based language model designed for Private Cloud Compute. These models are designed to perform a wide range of tasks efficiently, accurately, and responsibly. This report describes the model architecture, the data used to train the model, the training process, how the models are optimized for inference, and the evaluation results. We highlight our focus on Responsible AI and how the principles are applied throughout the model development.

Jean-Daniel Fekete, Yifan Hu, Dominik Moritz et al.

The rapid advancement of AI is transforming human-centered systems, with profound implications for human-AI interaction, human-data interaction, and visual analytics. In the AI era, data analysis increasingly involves large-scale, heterogeneous, and multimodal data that is predominantly unstructured, as well as foundation models such as LLMs and VLMs, which introduce additional uncertainty into analytical processes. These shifts expose persistent challenges for human-data interactive systems, including perceptually misaligned latency, scalability constraints, limitations of existing interaction and exploration paradigms, and growing uncertainty regarding the reliability and interpretability of AI-generated insights. Responding to these challenges requires moving beyond conventional efficiency and scalability metrics, redefining the roles of humans and machines in analytical workflows, and incorporating cognitive, perceptual, and design principles into every level of the human-data interaction stack. This paper investigates the challenges introduced by recent advances in AI and examines how these developments are reshaping the ways users engage with data, while outlining limitations and open research directions for building human-centered AI systems for interactive data analysis in the AI era.

Tobias Stähle, Péter Ferenc Gyarmati, Thilo Spinner et al.

The rise of AI agents introduces a fundamental shift in Visual Analytics (VA), in which agents act as a new user group. Current agentic approaches - based on computer vision and raw DOM access - fail to perform VA tasks accurately and efficiently. This paper introduces the Visual Analytics Context Protocol (VACP), a framework designed to make VA applications "agent-ready" that extends generic protocols by explicitly exposing application state, available interactions, and mechanisms for direct execution. To support our context protocol, we contribute a formal specification of AI agent requirements and knowledge representations in VA interfaces. We instantiate VACP as a library compatible with major visualization grammars and web frameworks, enabling augmentation of existing systems and the development of new ones. Our evaluation across representative VA tasks demonstrates that VACP-enabled agents achieve higher success rates in interface interpretation and execution compared to current agentic approaches, while reducing token consumption and latency. VACP closes the gap between human-centric VA interfaces and machine perceivability, ensuring agents can reliably act as collaborative users in VA systems.

Donghao Ren, Fred Hohman, Halden Lin et al. · apple-ml, cmu

Embedding projections are popular for visualizing large datasets and models. However, people often encounter "friction" when using embedding visualization tools: (1) barriers to adoption, e.g., tedious data wrangling and loading, scalability limits, no integration of results into existing workflows, and (2) limitations in possible analyses, without integration with external tools to additionally show coordinated views of metadata. In this paper, we present Embedding Atlas, a scalable, interactive visualization tool designed to make interacting with large embeddings as easy as possible. Embedding Atlas uses modern web technologies and advanced algorithms -- including density-based clustering, and automated labeling -- to provide a fast and rich data analysis experience at scale. We evaluate Embedding Atlas with a competitive analysis against other popular embedding tools, showing that Embedding Atlas's feature set specifically helps reduce friction, and report a benchmark on its real-time rendering performance with millions of points. Embedding Atlas is available as open source to support future work in embedding-based analysis.

Alex Oesterling, Donghao Ren, Yannick Assogba et al.

Safety policies define what constitutes safe and unsafe AI outputs, guiding data annotation and model development. However, annotation disagreement is pervasive and can stem from multiple sources such as operational failures (annotators misunderstand or misexecute the task), policy ambiguity (policy wording leaves room for interpretation), or value pluralism (different annotators hold different perspectives on safety). Distinguishing these sources matters. For example, operational failures call for quality control, ambiguity calls for policy clarification, and pluralism calls for deliberation about incorporating diverse perspectives. Yet understanding why annotators disagree is difficult. Directly asking annotators for their reasoning is costly, substantially increasing annotation burden, and can be unreliable for both human and LLM annotators as self-reported reasoning often fails to reflect actual decision processes. We introduce Annotator Policy Models (APMs), interpretable models that learn annotators' internal safety policies from labeling behavior alone, making annotator reasoning visible and comparable without additional annotation effort. We validate that APMs accurately model annotator safety policy (>80% accuracy), faithfully predict responses to counterfactual edits, and recover known policy differences in controlled settings. Applying APMs to LLM and human annotations, we demonstrate two core applications: (1) surfacing policy ambiguity by revealing how annotators interpret safety instructions differently, and (2) surfacing value pluralism by uncovering systematic differences in safety priorities across demographic groups. Together, these capabilities support more targeted, transparent, and inclusive safety policy design.

Nikhil Prakash, Donghao Ren, Dominik Moritz et al.

Prior studies investigating the internal workings of LLMs have uncovered sparse subnetworks, often referred to as circuits, that are responsible for performing specific tasks. Additionally, it has been shown that model performance improvement through fine-tuning often results from the strengthening of existing circuits in the model. Taken together, these findings suggest the possibility of intervening directly on such circuits to make precise, task-targeted updates. Motivated by these findings, we propose a novel method called Constructive Circuit Amplification which identifies pivotal tokens from model reasoning traces as well as model components responsible for the desired task, and updates only those components. Applied to mathematical reasoning, it improves accuracy by up to +11.4% across multiple models while modifying as little as 1.59% of model components, with minimal impact on other abilities as measured by MMLU, TriviaQA, and TruthfulQA. These results demonstrate that targeted capabilities can be reliably enhanced by selectively updating a sparse set of model components.

Fred Hohman, Chaoqun Wang, Jinmook Lee et al. · apple-ml, cmu

On-device machine learning (ML) moves computation from the cloud to personal devices, protecting user privacy and enabling intelligent user experiences. However, fitting models on devices with limited resources presents a major technical challenge: practitioners need to optimize models and balance hardware metrics such as model size, latency, and power. To help practitioners create efficient ML models, we designed and developed Talaria: a model visualization and optimization system. Talaria enables practitioners to compile models to hardware, interactively visualize model statistics, and simulate optimizations to test the impact on inference metrics. Since its internal deployment two years ago, we have evaluated Talaria using three methodologies: (1) a log analysis highlighting its growth of 800+ practitioners submitting 3,600+ models; (2) a usability survey with 26 users assessing the utility of 20 Talaria features; and (3) a qualitative interview with the 7 most active users about their experience using Talaria.

Ethan Li, Anders Boesen Lindbo Larsen, Chen Zhang et al. · apple-ml, cmu

We introduce two multilingual, multimodal foundation language models that power Apple Intelligence features across Apple devices and services: i a 3B-parameter on-device model optimized for Apple silicon through architectural innovations such as KV-cache sharing and 2-bit quantization-aware training; and ii a scalable server model built on a novel Parallel-Track Mixture-of-Experts PT-MoE transformer that combines track parallelism, mixture-of-experts sparse computation, and interleaved global-local attention to deliver high quality with competitive cost on Apple's Private Cloud Compute platform. Both models are trained on large-scale multilingual and multimodal datasets sourced via responsible web crawling, licensed corpora, and high-quality synthetic data, then further refined with supervised fine-tuning and reinforcement learning on a new asynchronous platform. The resulting models support several additional languages while understanding images and executing tool calls. In public benchmarks and human evaluations, both the server model and the on-device model match or surpass comparably sized open baselines. A new Swift-centric Foundation Models framework exposes guided generation, constrained tool calling, and LoRA adapter fine-tuning, allowing developers to integrate these capabilities with a few lines of code. The latest advancements in Apple Intelligence models are grounded in our Responsible AI approach with safeguards like content filtering and locale-specific evaluation, as well as our commitment to protecting our users' privacy with innovations like Private Cloud Compute.

Kushin Mukherjee, Donghao Ren, Dominik Moritz et al. · apple-ml, cmu

Multimodal vision-language models (VLMs) continue to achieve ever-improving scores on chart understanding benchmarks. Yet, we find that this progress does not fully capture the breadth of visual reasoning capabilities essential for interpreting charts. We introduce EncQA, a novel benchmark informed by the visualization literature, designed to provide systematic coverage of visual encodings and analytic tasks that are crucial for chart understanding. EncQA provides 2,076 synthetic question-answer pairs, enabling balanced coverage of six visual encoding channels (position, length, area, color quantitative, color nominal, and shape) and eight tasks (find extrema, retrieve value, find anomaly, filter values, compute derived value exact, compute derived value relative, correlate values, and correlate values relative). Our evaluation of 9 state-of-the-art VLMs reveals that performance varies significantly across encodings within the same task, as well as across tasks. Contrary to expectations, we observe that performance does not improve with model size for many task-encoding pairs. Our results suggest that advancing chart understanding requires targeted strategies addressing specific visual reasoning gaps, rather than solely scaling up model or dataset size.

Donghao Ren, Fred Hohman, Dominik Moritz · apple-ml, cmu

Interactive visualization of embedding projections is a useful technique for understanding data and evaluating machine learning models. Labeling data within these visualizations is critical for interpretation, as labels provide an overview of the projection and guide user navigation. However, most methods for producing labels require clustering the points, which can be computationally expensive as the number of points grows. In this paper, we describe an efficient clustering approach using kernel density estimation in the projected 2D space instead of points. This algorithm can produce high-quality cluster regions from a 2D density map in a few hundred milliseconds, orders of magnitude faster than current approaches. We contribute the design of the algorithm, benchmarks, and applications that demonstrate the utility of the algorithm, including labeling and summarization.

Angie Boggust, Donghao Ren, Yannick Assogba et al. · apple-ml, cmu

Automated interpretability aims to translate large language model (LLM) features into human understandable descriptions. However, these natural language feature descriptions are often vague, inconsistent, and require manual relabeling. In response, we introduce semantic regexes, structured language descriptions of LLM features. By combining primitives that capture linguistic and semantic feature patterns with modifiers for contextualization, composition, and quantification, semantic regexes produce precise and expressive feature descriptions. Across quantitative benchmarks and qualitative analyses, we find that semantic regexes match the accuracy of natural language while yielding more concise and consistent feature descriptions. Moreover, their inherent structure affords new types of analyses, including quantifying feature complexity across layers, scaling automated interpretability from insights into individual features to model-wide patterns. Finally, in user studies, we find that semantic regex descriptions help people build accurate mental models of LLM feature activations.

Alex Bäuerle, Ángel Alexander Cabrera, Fred Hohman et al.

Interfaces for machine learning (ML), information and visualizations about models or data, can help practitioners build robust and responsible ML systems. Despite their benefits, recent studies of ML teams and our interviews with practitioners (n=9) showed that ML interfaces have limited adoption in practice. While existing ML interfaces are effective for specific tasks, they are not designed to be reused, explored, and shared by multiple stakeholders in cross-functional teams. To enable analysis and communication between different ML practitioners, we designed and implemented Symphony, a framework for composing interactive ML interfaces with task-specific, data-driven components that can be used across platforms such as computational notebooks and web dashboards. We developed Symphony through participatory design sessions with 10 teams (n=31), and discuss our findings from deploying Symphony to 3 production ML projects at Apple. Symphony helped ML practitioners discover previously unknown issues like data duplicates and blind spots in models while enabling them to share insights with other stakeholders.

Jochen Görtler, Fred Hohman, Dominik Moritz et al.

The confusion matrix, a ubiquitous visualization for helping people evaluate machine learning models, is a tabular layout that compares predicted class labels against actual class labels over all data instances. We conduct formative research with machine learning practitioners at Apple and find that conventional confusion matrices do not support more complex data-structures found in modern-day applications, such as hierarchical and multi-output labels. To express such variations of confusion matrices, we design an algebra that models confusion matrices as probability distributions. Based on this algebra, we develop Neo, a visual analytics system that enables practitioners to flexibly author and interact with hierarchical and multi-output confusion matrices, visualize derived metrics, renormalize confusions, and share matrix specifications. Finally, we demonstrate Neo's utility with three model evaluation scenarios that help people better understand model performance and reveal hidden confusions.

Hyeok Kim, Ryan Rossi, Abhraneel Sarma et al.

Authors often transform a large screen visualization for smaller displays through rescaling, aggregation and other techniques when creating visualizations for both desktop and mobile devices (i.e., responsive visualization). However, transformations can alter relationships or patterns implied by the large screen view, requiring authors to reason carefully about what information to preserve while adjusting their design for the smaller display. We propose an automated approach to approximating the loss of support for task-oriented visualization insights (identification, comparison, and trend) in responsive transformation of a source visualization. We operationalize identification, comparison, and trend loss as objective functions calculated by comparing properties of the rendered source visualization to each realized target (small screen) visualization. To evaluate the utility of our approach, we train machine learning models on human ranked small screen alternative visualizations across a set of source visualizations. We find that our approach achieves an accuracy of 84% (random forest model) in ranking visualizations. We demonstrate this approach in a prototype responsive visualization recommender that enumerates responsive transformations using Answer Set Programming and evaluates the preservation of task-oriented insights using our loss measures. We discuss implications of our approach for the development of automated and semi-automated responsive visualization recommendation.

Hyeok Kim, Dominik Moritz, Jessica Hullman

Increased access to mobile devices motivates the need to design communicative visualizations that are responsive to varying screen sizes. However, relatively little design guidance or tooling is currently available to authors. We contribute a detailed characterization of responsive visualization strategies in communication-oriented visualizations, identifying 76 total strategies by analyzing 378 pairs of large screen (LS) and small screen (SS) visualizations from online articles and reports. Our analysis distinguishes between the Targets of responsive visualization, referring to what elements of a design are changed and Actions representing how targets are changed. We identify key trade-offs related to authors' need to maintain graphical density, referring to the amount of information per pixel, while also maintaining the "message" or intended takeaways for users of a visualization. We discuss implications of our findings for future visualization tool design to support responsive transformation of visualization designs, including requirements for automated recommenders for communication-oriented responsive visualizations.

Connie Ye, Lukas Hermann, Nur Yildirim et al.

NASA JPL scientists working on the micro x-ray fluorescence (microXRF) spectroscopy data collected from Mars surface perform data analysis to look for signs of past microbial life on Mars. Their data analysis workflow mainly involves identifying mineral compounds through the element abundance in spatially distributed data points. Working with the NASA JPL team, we identified pain points and needs to further develop their existing data visualization and analysis tool. Specifically, the team desired improvements for the process of creating and interpreting mineral composition groups. To address this problem, we developed an interactive tool that enables scientists to (1) cluster the data using either manual lasso-tool selection or through various machine learning clustering algorithms, and (2) compare the clusters and individual data points to make informed decisions about mineral compositions. Our preliminary tool supports a hybrid data analysis workflow where the user can manually refine the machine-generated clusters.

Aoyu Wu, Yun Wang, Xinhuan Shu et al.

Visualizations themselves have become a data format. Akin to other data formats such as text and images, visualizations are increasingly created, stored, shared, and (re-)used with artificial intelligence (AI) techniques. In this survey, we probe the underlying vision of formalizing visualizations as an emerging data format and review the recent advance in applying AI techniques to visualization data (AI4VIS). We define visualization data as the digital representations of visualizations in computers and focus on data visualization (e.g., charts and infographics). We build our survey upon a corpus spanning ten different fields in computer science with an eye toward identifying important common interests. Our resulting taxonomy is organized around WHAT is visualization data and its representation, WHY and HOW to apply AI to visualization data. We highlight a set of common tasks that researchers apply to the visualization data and present a detailed discussion of AI approaches developed to accomplish those tasks. Drawing upon our literature review, we discuss several important research questions surrounding the management and exploitation of visualization data, as well as the role of AI in support of those processes. We make the list of surveyed papers and related material available online at ai4vis.github.io.

Mary Beth Kery, Donghao Ren, Fred Hohman et al.

We aim to increase the flexibility at which a data worker can choose the right tool for the job, regardless of whether the tool is a code library or an interactive graphical user interface (GUI). To achieve this flexibility, we extend computational notebooks with a new API mage, which supports tools that can represent themselves as both code and GUI as needed. We discuss the design of mage as well as design opportunities in the space of flexible code/GUI tools for data work. To understand tooling needs, we conduct a study with nine professional practitioners and elicit their feedback on mage and potential areas for flexible code/GUI tooling. We then implement six client tools for mage that illustrate the main themes of our study findings. Finally, we discuss open challenges in providing flexible code/GUI interactions for data workers.

Dominik Moritz, Danyel Fisher

Data analysts often need to work with multiple series of data---conventionally shown as line charts---at once. Few visual representations allow analysts to view many lines simultaneously without becoming overwhelming or cluttered. In this paper, we introduce the DenseLines technique to calculate a discrete density representation of time series. DenseLines normalizes time series by the arc length to compute accurate densities. The derived density visualization allows users both to see the aggregate trends of multiple series and to identify anomalous extrema.

Bahador Saket, Dominik Moritz, Halden Lin et al.

In this paper, we describe a research agenda for deriving design principles directly from data. We argue that it is time to go beyond manually curated and applied visualization design guidelines. We propose learning models of visualization design from data collected using graphical perception studies and build tools powered by the learned models. To achieve this vision, we need to 1) develop scalable methods for collecting training data, 2) collect different forms of training data, 3) advance interpretability of machine learning models, and 4) develop adaptive models that evolve as more data becomes available.